Machine Learning in Manufacturing – gimmick or practicable?

"Machines will be capable, within 20 years, of doing any work a man can do," said Nobel prize winner, Herbert Simon, in the 1960s. Obviously, he was a little hasty. When prominent physicist, Stephen Hawking, warned two years ago "Computers will overtake humans within 100 years", many experts agreed that this is a realistic statement, considering the current technological progress. Additionally, the latest developments in Machine Learning support these assessments.

It is certain that Machine Learning is a big issue for industries, regardless of their sector. Companies such as Apple, Google and Facebook have already invested in startups dealing with the subject. However, how the trending technology can be used and how far it will go to replace humans in the future is still under discussion.

Current discussions are dominated by gimmicks of the technology, but when used in the proper sense, Machine Learning is meant to support useful processes. Primarily, it is about learning from data, looking for patterns and thereupon building predictions to predetermined questions. The machine learns from existing data pools or patterns. Although humans teach the machines, they no longer control them.

Machine Learning in Manufacturing

Machine Learning has particularly great potential for optimizing processes in machine and plant engineering. Every day, machines and plants produce huge quantities of useful data. Therefore, it is required that intelligent systems collecting data are implemented in these fields. It is not yet clear how these data masses - often called Big Data - can be transformed into useful information; however, many possible applications have already been planned.

It could benefit production planning, when the machines learn from the past. The planned values for production planning are often incorrect and do not reflect reality. For example, 400 minutes could be planned for a machine´s work step, although the machine usually needs around 100 minutes for the process. This shows how utilization of resources cannot be optimally reached. Although these production processes will become trouble-free, due to the major time differences, the machines´ inactivity times will also increase.

Another example is supplier management. Sometimes production managers are able to evaluate their suppliers’ performance based on data, and in the worst case, reprimand them. However, the supplier could still not deliver on time. In this case, it would be more efficient to consider delays directly in the planning. The machine would learn from the past and enable optimal processes. It would “know” in detail, when it has to be serviced or when a part needs to be exchanged, because defects from abrasion can be predicted in advance. The learned model could then state a specific time for replacements, avoiding the need for unnecessary changes at regular intervals.

Closing thoughts

The development and interest in Machine Learning is increasing. The number of possible application fields is endless. There is already high potential for the technology in machine and plant engineering as there is a lot of available data due to the use of intelligent planning systems. Well-maintained data is required for reasonable predictions. Great thinkers already predicted the advancement of Machine Learning a long time ago. However, machines cannot yet think, and human support is still needed to optimize production processes.

How do you see the future of Machine Learning in Manufacturing?

 

---

---

You may also like